Add stress/sleep ML train harness and reproduce pack

[Newbbsss]

Summary

• Add end-to-end stress tip + sleep Rest/stage ML reproduce pack: Tools/stress_sleep_train.py, ml_engine_train.py, sleep_stage_eval.py, calibrate_whoop_noop.py, ingest_export.ps1.
• Check in golden labels/manifests (decoded screenshot packs) + factor docs + docs/STRESS_SLEEP_ML_REPRO.md so someone else can recreate results without Fold live ADB.
• Desk (Host desk) CPU train run completed; accuracy_valid stays false until documented N gates (honest underfit with 1 tip@clock pair).

Reproduce stress + sleep

python Tools\stress_sleep_train.py --as-of 2026-07-14
python Tools\ml_engine_train.py --labels pairing-logs\whoop-app-labels.jsonl --as-of 2026-07-14
python Tools\sleep_stage_eval.py --synthetic
# after PhysioNet sleep-accel:
# python Tools\sleep_stage_eval.py --sleep-accel-dir pairing-logs\datasets\sleep-accel

See docs/STRESS_SLEEP_ML_REPRO.md.

Artifacts

• pairing-logs/stress-sleep-train-status.json
• pairing-logs/ml-stress-tip-weights.json
• pairing-logs/sleep-stage-eval.json
• pairing-logs/whoop-app-labels.jsonl (adb + screenshot decode)
• Decoded manifests: pairing-logs/exports/20260712-Noop_stresd/, .../20260713-Noop_mg/

Notes / blockers

• Fold plugged in but powered OFF this session: no bank pull, tip@clock live, or APK live-edit. Device verification deferred.
• Primary tip@clock: WHOOP 1.4 MEDIUM vs NOOP 1.8 @ 22:14 on 2026-07-13 (δ +0.4). Sleep: WHOOP 64% vs NOOP Rest 73 with Deep 2%/REM 0% (REST-WEIGHT).
• No MAIN APK bump: docs/ML only.

AI assistance

Prepared with Cursor agent assistance.

[Newbbsss]

Closing: this PR targeted the Newbbsss fork by mistake. The lean stress/sleep ML reproduce pack now targets upstream here:

ryanbr#418

[Newbbsss]

Superseded by ryanbr#418 (base: ryanbr/noop main).